Study Limitations

A number of weaknesses of this study must be noted. The first is the cross-sectional nature of the analysis. Longitudinal cognitive measures were not available, and in their absence, this study was not able to provide evidence for a causal association between the BE and

condition. Also, the MESA data were not collected to study the BE and cognition specifically, and therefore, the pre-existing BE and cognitive measures may not be ideally suited for that

particular purpose. In addition to the measures of PA and trust of neighbors, future studies should investigate whether factors such as air pollution and stress are mediators of the BE- cognition associations. Furthermore, longitudinal and lifecourse measures of the BE will be needed to assess whether early-life, mid-life, late-life BE exposures, or a combination of the three, are the most important to consider in relation to cognition in older adults. Lifelong residential mobility but also residential moves at older ages will be important to consider in future studies, particularly if individuals move between substantially different BEs over time. Although the neighborhood BE was hypothesized to be the most pertinent environment to examine with regards to older adults, because they are typically retired and spend more time closer to home, other environments may also be important to consider. For instance, it may be important to examine the environment surrounding workplaces among older adults who continue to work or volunteer, but also neighborhoods adjacent to home neighborhoods (e.g., if a low SES neighborhood is surrounded by many high or low SES neighborhoods).

It is possible that my results were biased due to sample selection, self-selection into neighborhoods, or misclassification. The MESA sample was derived from population-based methods, which is a strength, but the six US sites may represent different BEs than other regions of the US; thus, threatening external validity. Sample selection of the analytic sample may have also biased the results, if characteristics of the excluded individuals are associated with both the BE and cognitive measures. The Exam 1 participants that were excluded had lower levels of education, were less often married, were more often of non-white race, and had lower family income, and therefore, excluding these participants may have biased the results. While the possibility of self-selection into neighborhoods is frequently mentioned as a weakness of many BE and health studies, there is some evidence that this may not cause significant bias152,

although this should be examined specifically for studies of the BE and cognition. Finally, if the BE or cognitive measures did not accurately measure the desired constructs, then the results may have been affected by misclassification bias. For example, this would have occurred if the measure of distance to nearest train stop did not accurately and reliably measure transit accessibility or if the DS test was not a valid and reliable measure of processing speed. 5.3. Future studies

In my study, I examine a small portion of the potential associations between the BE and cognition, as outlined in my conceptual model (Figure 1.2). Although I included measures of accessibility, land use, transit availability, and population density, there are many other measures that could be explored in future studies, including measures of park and green space, housing, and design (e.g., condition of public spaces). In addition, future studies could investigate the association between cognition and additional measures of accessibility, land use (e.g., entropy), transportation features (e.g., sidewalk availability), and density (e.g., housing density). The cognitive measures used in this study included a brief cognitive measure, and measures of short- term and working memory and processing speed. Future studies would benefit by examining better overall cognitive measures compared to the CASI, as well as measures of other cognitive domains that may be differentially affected by BE exposures, such as memory, executive function, and visual-spatial function.

Similarly, future studies could explore additional moderators and mediators of the BE- cognition association. Plausible moderators that were not examined in my study include age, time living in the neighborhood, and physical, cognitive, and health status (e.g., disability). Certain aspects of the BE may have a stronger influence on cognition among individuals who are older, who have lived less or more time in the neighborhood, or who have certain physical or

health conditions. For example, older individuals and individuals with physical disabilities may be more affected by neighborhood sidewalks in poor condition. It is possible that in addition to individual-level moderators, two or more neighborhood-level characteristics have additive or multiplicative effects when an individual is exposed to them. For instance, individuals living in neighborhoods of with high population density and with higher levels of psychosocial hazards (e.g., graffiti, broken windows, loitering) may have worse cognition than individuals living in neighborhoods with none or only one of these characteristics. This dissertation study was not aimed at assessing mediators, and therefore, there are numerous potential mediators that future studies can investigate including air pollution exposures; quality of life measures; mental health measures of stress, anxiety, and depression; social measures of engagement, isolation,

neighboring support, and trust in neighbors; and health-related behaviors including leisure-time and utilitarian PA, diet, and health care access.

New studies can build upon the methods in this study by using longitudinal BE and cognitive measures, to provide evidence for a causal association.192 In particular, lifecourse approaches would help tease apart whether, in addition to early- or mid-life exposures and risk factors that are unrelated to the BE, long-term BE exposures are associated with differences in cognition in older adults. Studies are also needed to evaluate the objective and perceived neighborhood boundaries among older adults, and whether these definitions differ based on the BE and cognitive measures being examined. For instance, land use mix may be most important in the area immediately surrounding an individual’s home, whereas the availability of green space could be important up to one mile from the home. A comparison of various definitions of the same BE features may help to clarify and expand upon some of the observed relationships in this study. As an example, future studies could investigate how different measures of transit

accessibility (density of transit stops, distance to nearest stops, and accessibility of transit stops to the rest of the region) are associated with cognition in older adults.